Dynamic Soft Real-Time Scheduling with Preemption ...downloads.hindawi.com/journals/ijdmb/2019/5284968.pdf · works as so real-time scheduling. e streaming packet constructing priority

Research ArticleDynamic Soft Real-Time Scheduling with PreemptionThreshold for Streaming Media

WenleWang 1 YuanWang1 JiangYan Dai 2 and Zhonghua Cao3

1School of Software Jiangxi Normal University Nanchang 330022 Jiangxi China2School of Computer Engineering Weifang University Weifang 261061 Shandong China3School of Software amp Communication Engineering Jiangxi University of Finance and Economics Nanchang 330038 Jiangxi China

Correspondence should be addressed to JiangYan Dai daijyan163com

Received 21 September 2018 Revised 19 November 2018 Accepted 3 December 2018 Published 1 January 2019

Guest Editor Kai Wang

Copyright copy 2019 Wenle Wang et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Over the last decades the advancements in networking technology and new multimedia devices have motivated the researchon efficient video streaming mechanisms under wireless We consider combing soft real-time video streaming schedulingwith threshold to minimize the ineffective preemption Based on the value density and urgency of soft real-time task thedynamic scheduling with preemption threshold strategy (DSPT) is proposed in the paper By analyzing the response time andpreemption relationship of tasks the preemption thresholds are assigned Simulation results show that the DSPT strategy achievesimprovements about success rate delay time and benefit of the system

1 Introduction

With the advancements in networking technology and newmultimedia devices video streaming mechanisms undervarious network and devices have attracted more and moreattention Nowadays multimedia communications over theinternet especially wireless network are the necessarily partof modern life Many researchers have studied on videostreamingmanagement over wireless including video coding[1ndash4] network coding [5ndash8] and wireless communication[9ndash11] In wireless networks video streaming applicationsare almost the largest consumer of mobile wireless dataso scheduling video streaming in wireless networks is veryimportant

Video streaming scheduling in wireless networks hasbeen developed for recent years such as given quality onservice mechanisms [12ndash16] and real-time adaptive [17ndash19]Many of video streaming scheduling studies are about hardreal-time [20ndash22] that requires all the packets completebefore the deadline otherwise the packets will be consideredinvalid and dropped However for some actual video stream-ing media application the packets of the transmission withdelay can still bring a certain service and these packets can be

forwarded and represented by soft real-time tasks The softreal-time task model with value function is also applicableto video streaming with delay [23 24] For soft real-timescheduling task executing under delay-constrained can bringgoodput and benefit

Generally real-time scheduling can be divided into pre-emptive scheduling and non-preemptive scheduling accord-ing to whether the task can preempt each other or notIn a dynamic network the preemptive method is moreapplicable However arbitrary preemption between tasks willaffect the scheduling performanceTherefore considering thedrawbacks of the two methods a compromise approach isutilized to limit the arbitrary or invalid preemption by settingpreemption threshold Furthermore the limited preemptionscheduling strategy is a correct choice for improving theperformance

Soft real-time task model is conducted for the videostreaming under wirelessWe propose a soft real-time thresh-old preemption scheduling strategy by combining value den-sity and urgency for video streaming in the paper Simulationresults show that the strategy can improve the goodputreduce the delay and increase the benefit comparing with theEDF and LSF methods

HindawiInternational Journal of Digital Multimedia BroadcastingVolume 2019 Article ID 5284968 7 pageshttpsdoiorg10115520195284968

2 International Journal of Digital Multimedia Broadcasting

In this paper we only consider the CPU in the systemscheduling and ignore the time caused by context switchingWe study the video streaming transmission problem ofgateway devices in wireless network at the specific level ofpackets of transport tiers

Organization The structure of the paper is organizedas follows Section 2 describes the related work Section 3designs the structure of DSPT strategy Section 4 definesthe task model and problem analysis in detail Section 5 ispriority and preemption thresholds description The detailsof scheduling based on preemption threshold are introducedin Section 6 The simulation setup and results are givenin Section 7 Finally the conclusions and future work arediscussed

2 Related Work

In wireless networks the scheduling strategy should ensurehigh quality For this purpose lots of studies were proposedThe works in [1 2] enabled high-quality HTTP streamingin which video pre-encoded into segments that can onlystart to play if all the packet have received However muchpacket lost for video application in the above work Theliterature [3] optimized the dynamic HTTP live streamingservice by segment adaption Some literatures proposedvideo streaming scheduling using video coding with networkcoding For example the literatures [5 6] optimized videostreaming by combining video coding and network codingfor P2P network and distributed network respectively How-ever it might suffer from bandwidth inefficiencies becauseof unrecoverable packets The work [7] proposed streamingscheduling for multiple server by Markov decision processesbut it had high computational complexity In [9] the authorsdesigned a crosslayer optimization scheme for dynamicwireless video streaming However these methods cannotcontain the packet urgency for streaming control

With the emergence of various networks the researchon heterogeneous wireless networks had been motivatedThe literatures [11 19] studied the concurrent multipathtransfer of the capacity-limited heterogeneous wireless plat-form Another work [17] proposed a real-time adaptive videostreaming delivermethod overmultiple wireless networks In[23 24] delay-constrained video transmission with multipleinterface in heterogeneous network was proposed

Some researchers considered video streaming as hardreal-time that is the streaming packet missing the deadlinewill be considered invalid For example the literature [20]proposed a feedback-based control approach under hardreal-time workload to improve admission rate In literatures[18 21] the author studied in hard real-time video streamallocation to improve the utilization of video decodingNevertheless most video streaming data meets the charac-teristics of soft real-time task so delay-constrained of videois considered acceptable in applications

3 DSPT Detail Design

We consider video streaming deliver over wireless net-works as soft real-time scheduling The streaming packet

constructing priority function

middot preemption threshold selection middot response time analysis

scheduling based on preemption threshold

task model and analysis middot value density analysis

middot urgency analysis

task schedulability analysis Scheduler

execute

1 i

i

j

Figure 1 The flow diagram of DSPT strategy

is expressed as task and its quality and urgency are themetrics in scheduling The taskrsquos delay will affect the qualityTherefore we introduced the value to measure the qualityResponse time analysis (RTA) is used to test the schedula-bility of real-time task and the preemption threshold-basedscheduling approach is to reduce invalid preemption amongtasks The scheduling algorithm can effectively reduce thedelay and improve the success rate and benefit The flowdiagram of DSPT strategy is shown in Figure 1 In this paperwe assume that the value function 119881119894(119905) of the soft real-time task 119879119894 keeps unchanged in the interval [119886119894 119889119894) anddecreases linearly in the interval While executing in theinterval [119889119894 119888119903119894) the task 119879119894 is not allowed to be preemptedin order to protect the execution of the task

Contribution 1 We construct the priority assignment func-tion combining value density and urgency improving thesuccess and quality of task soft real-time scheduling

Contribution 2 We propose the task scheduling strategyDSPT based on preemption threshold which can obtainbetter performance

4 Task Model and Analysis

Considering video streaming with delay-constrained underwireless we conduct soft real-time task model in this section

41 Soft Real-Time Task Model We refer to a task as anaperiodic task model and the task is soft real-time Thissection provides the definitions of soft real-time task modelused in the paper

Task 119879119894 An aperiodic task 119879119894 is defined as a triple 119879119894 =119886119894 119890119905119894 119889119894 119888119903119894 where 119886119894119890119905119894119889119894 and 119888119903119894 are the arrival timethe estimated execution time the relative deadline and thecritical time of task 119879119894

Having executed time ℎ119905119894 ℎ119905119894 denotes that the task hasbeen executing ℎ119905119894 units at time 1199050

Required execution time 119903119905119894 119903119905119894 denotes the remainingexecution time of task 119879119894 Thus it meets 119903119905119894 = 119890119905119894 minus ℎ119905119894

Executable time 119897119905119894 119897119905119894 is the executable time of the task 119879119894prior to its deadline 119889119894 where at the time 1199050(1199050 lt 119889119894)

International Journal of Digital Multimedia Broadcasting 3

Soft Executable time 119904119897119905119894 When missed deadline 119889119894 taskcan continue to execute until 119888119903119894 119904119897119905119894 is the executable timeprior to the critical time 119888119903119894 that is 119904119897119905119894 = 119888119903119894 minus 1199050

Slack time 119904119905119894The slack time 119904119905119894 denotes that the time unitafter finish execution until deadline 119889119894 thus 119904119905119894 = 119897119905119894 minus 119903119905119894119904119905119894 remains stable when the task 119879119894 is in execution and 119904119905119894decreases when the task 119879119894 is preempted Once 119904119905119894 lt 0 thetask 119879119894 cannot finish before 119889119894

Delay Time 119889119905119894 The delay time 119889119905119894 denotes the finish timeof task 119879119894 after 119889119894 At the time 1199050(119889119894 lt 1199050 le 119888119903119894) it meets 119889119905119894 =1199050 minus 119889119894

Priority 119901119903119894 119901119903119894 is defined as the priority of task 119879119894 Thelarger 119901119903119894rsquos value the higher 119879119894rsquos priority

Threshold 120579119894 120579119894 is the preemption threshold of task 119879119894where 120579119894 ge 119901119903119894

Value function 119881119894(119905) 119881119894(119905) defines the value of the systemfor executing 119879119894 at time 119905 where 119905 isin [119886119894 119888119903119894) The task 119879119894 rsquosvalue will change when 119879119894 keeps executing and the function119881119894(119905) is dynamic 119881119894(119905) will be analyzed in detail in Section 4

Value density function119881119863119894(119905)119881119863119894(119905) is the ratio between119881119894(119905) and required execution time 119903119905119894 that is 119881119863119894(119905) =119881119894(119905)119903119905119894 Obviously 119881119863119894(119905) is growing with the decreasing of11990311990511989442 Value and Value Density Function of Soft Real-Time TaskAccording to assumption the value function 119881119894(119905) of the softreal-time task 119879119894 at time 119905 is defined as

119881119894 (119905) = 119881119894 119905 lt 119889119894119888119903119894 minus 119905119888119903119894 minus 119889119894 times 119881119894 119889119894 le 119905 lt 119888119903119894 (1)

From (1) we can see that when 119905 lt 119889119894 119881119894(119905) is stable at 119881119894however when 119889119894 le 119905 lt 119888119903119894 119881119894(119905) is linearly diminishing until0

Value density function 119881119863119894(119905) is denoted as

119881119863119894 (119905) =

119881119894119903119905119894 119905 lt 119889119894119888119903119894 minus 119905119888119903119894 minus 119889119894 times

119881119894119903119905119894 119889119894 le 119905 lt 119888119903119894 (2)

We can see that 119881119863119894(119905) is not only concerned with 119881119894(119905) butalso related to 119903119905119894 According to the definition 119903119905119894 = 119890119905119894 minus ℎ119905119894Aswe knownwhen the task119879119894 is executing 119903119905119894 decreases withthe growing of the time 119905 Andwhen the task119879119894 is blocked the119903119905119894 remains unchangedThediscussion of119881119863119894(119905) is as follows

(1) 119905 lt 119889119894 Within this interval 119881119863119894(119905) = 119881119894119903119905119894 When 119879119894is executing 119881119863119894(119905) grows with the decreasing of 119903119905119894While 119879119894 is blocked 119881119863119894(119905) remains unchanged

(2) 119889119894 le 119905 lt 119888119903119894 In this interval 119881119863119894(119905) can be derived asfollows 119881119863119894(119905) = (119881119894119903119905119894) times ((119888119903119894 minus 119905)(119888119903119894 minus 119889119894))From the above expression the affect part of the119881119863119894(119905) is 119888119903119894 minus 119905119903119905119894 (3)

Because the task in scheduler can finish without anyinterference of any other task (119888119903119894 minus 119905)119903119905119894 gt 1 In thiscase if the task119879119894 keeps executing the molecular partand the denominator part of (3) decrease at the sametime Further analysis is needed

(1) At 1199050 time (3) is recorded as (119888119903119894minus1199050)119903119905119894 Whentask119879119894 has executedΔ119905 from 1199050 (3) is formulatedto (119888119903119894minus(1199050+Δ119905))(119903119905119894minusΔ119905) Because 119904119897119905119894119903119905119894 gt 1then (119888119903119894 minus (1199050 + Δ119905))(119903119905119894 minus Δ119905) gt (119888119903119894 minus 1199050)119903119905119894where Δ119905 gt 0 That is when task 119879119894 is executing119881119863119894(119905) grows with the decreasing of 119903119905119894

(2) When the task 119879119894 is blocked the molecular partof (3) decreases with the growth of 119905 whilethe denominator 119903119905119894 is kept unchanged Thus(3) decreases with the growth of 119905 ThereforeTheorem 1 can be obtained

Theorem 1 While the task keeps executing in the progress itincreases with time 11990543 Analysis of the Taskrsquos Urgency The more urgency of thetask is the sooner execution is required

Definition 2 At any 119905 the urgency119880119892119894(119905) of task 119879119894 is definedas 119880119892119894(119905) = 1119890119904119905119894

119880119892119894(119905) is affected by 119904119905119894 the discussion is as follows(1) When 119879119894 remains in execution 119904119905119894 does not change

with the growing of the time 119905 and then119880119892119894(119905) remainsunchanged

(2) When 119879119894 is blocked 119904119905119894 decreases with the growth of119905 which makes 119880119892119894(119905) increase5 Priority and Preemption Thresholds for SoftReal-Time Tasks

Task scheduling is driven by priority while task priorityfunction combines value density with urgency

51The Construction of Taskrsquos Priority Function Consideringthe urgency and the benefits of task task scheduling is basedon priorityOnly the completed task can gain value so relativeto the value density the urgency is more important In systemscheduling the first consideration is system success rate andthen is value brought by tasks It means that urgency is givenmuch greater weight than value density in priority functionTherefore we use exponential weighting to the urgencywhile using logarithmic weighting to the value density Thefunction of priority assignment can be rewritten to

119901119903119894 = 119880119892119894 (119905) times ln [1 + 119881119863119894 (119905)] (4)

Based on the above analysis of value density 119881119863119894(119905) andurgency 119880119892119894(119905) we can conclude that

(1) When 119879119894 executes at interval [119886119894 119889119894) 119901119903119894 satisfies119901119903119894 = 1119890 (119904119905119894) times ln [1 + 119881119894119903119905119894 ] (5)


Further analysis is needed

(1) According to Theorem 1 when 119879119894 keeps run-ning ln[1+119881119894119903119905119894] increases Because 119904119905119894 remainunchanged with time 119905 growing Therefore (5)keeps increasing

(2) Once 119879119894 is preempted 1119890119904119905119894 augments as time 119905grows while ln[1 + 119881119894119903119905119894] remain unchangedthen (5) keeps increasing Based on above 119901119903119894increases as time 119905 at interval [119886119894 119889119894)

(2) When 119879119894 executes at [119889119894 119888119903119894) its slack time is 119904119897119905119894 andsatisfies

119901119903119894 = 1119890119904119905119894 times ln [1 + 119888119903119894 minus 119905119888119903119894 minus 119889 times119881119894119903119905119894 ] (6)

(1) In the process of running119879119894 with the growing ofthe time 119905 ((119888119903119894minus119905)(119888119903119894minus119889119894))times(119881119894119903119905119894) increaseswhile 119904119897119905119894 remains unchanged and then (6) keepsincreasing

(2) When 119879119894 has been preempted with the growingof time 119905 119904119897119905119894 decreases and 1119890119904119897119905119894 augmentsaccordingly while ((119888119903119894 minus 119905)(119888119903119894 minus 119889119894)) times (119881119894119903119905119894)reduces Further discussion of Eq (6) is neededAt moment 119905 119904119897119905119894 = 119888119903119894 minus 119905 minus 119903119905119894 lt 119888119903119894 minus 119889119894After task 119879119894 has been executed with Δ119905 thereis 119901119903119894(Δ119905) = (1119890119888119903119894minus119905minus119903119905119894minusΔ119905) times ln[1 + ((119888119903119894 minus 119905 minusΔ119905)(119888119903119894minus119889119894))times(119881119894119903119905119894)] where 119888119903119894minus119905minus119903119905119894minusΔ119905 gt0 And 1119890119888119903119894minus119905minus119903119905119894minusΔ119905 is exponential growth inwhich growth rate is faster than ln[1 + ((119888119903119894 minus 119905minusΔ119905)(119888119903119894 minus 119889119894)) times (119881119894119903119905119894)]rsquos logarithmic descentrate significantly Therefore 119901119903119894(Δ119905) increaseswhen Δ119905 grows

52 Preemption Threshold Selection for Task The selectionof preemption thresholds directly affects the schedulingalgorithm We select preemption threshold by analyzing thetaskrsquos response time

Definition 3 Task 119879119894rsquos Effect Job Set is denoted as 119864119869119878119894During the execution of task 119879119894 all task of the sets canpreempt 119879119894 which are defined as Effect Job Set (119864119869119878119894)satisfying 119864119869119878119894 = 119879119898 | (119901119903119894119898 gt 119901119903119894119894and(119863119898 gt 119860 119894)and(119860119898 lt 119863119894)

Response time119877119905119894 of task119879119894 contains two parts estimatedexecution time 119890119905119894 and sum of the remaining execution timeof the effect job sets 119864119869119878119894

119877119905119894 = 119890119905119894 + sum119879119895isin119864119869119878119894

119903119905119895 (7)

By analyzing the response time of 119879119894 its threshold 120579119894 canbe obtained under the condition which can be scheduledThe procedure of preemption threshold selection is shown inAlgorithm 1 where 119879119898119886119909 is the highest priority task

Obviously the computational complexity of Algorithm 1is 119874(119899)

1 Calculate 120579119894 using (7)2 120579119894 = 1199011199031198943 WHILE (119877119905119894 le 119888119903119894) DO4 IF (120579119894 gt 119901119903119898119886119909)5 return not scheduled6 ELSE7 IF 119879119894 is executing and 119904119905119894 lt 1199041199051198981198861199098 120579119894 = 1199011199031198981198861199099 return 12057911989410 ELSE11 Calculate 119877119905119894 using (7)12 END IF13 END IF14 ENDWHILE15 return 120579119894

Algorithm 1 Compute 120579119894

6 Task Scheduling Strategy Based onPreemption Threshold

We propose a limited preemption scheduling strategy DSPTbased on preemption threshold which reduces the blockingtime of the tasks by restricting preemption

Definition 4 Sufficient and necessary conditions of 119901119903119894 gt 120579119895is that task 119879119894 can preempt task 119879119895

By Definition 4 if task 119879119894 preempts task 119879119895 there must be119901119903119894 gt 119901119903119895Theorem 5 If task 119879119894 and task 119879119895 are not preempted thefollowing should be satisfied (119901119903119894 le 120579119895) and (119901119903119895 le 120579119894)Proof Supposing that there are two tasks 119879119901 and 119879119902 whichdo not preempt each other ((119901119903119901 le 120579119902)and (((119901119903119902 le 120579119901)))) 997904rArr(119901119903119901 gt 120579119902) or (119901119903119902 gt 120579119901) is satisfied

There are two possible cases

(1) If 119901119903119901 gt 120579119902 task 119879119901 can preempt task 119879119902(2) If 119901119903119902 gt 120579119901 task 119879119901 may be preempted by task 119879119902Obviously tasks 119879119901 and 119879119902 can preempt each other in

contradiction with Definition 4ThereforeTheorem 5 can beproved

61 Scheduling Algorithm Based On Preemption ThresholdThe soft real-time dynamic task scheduling based on preemp-tion threshold (DSPT) called Algorithm 1 is to calculate thethreshold and determines the scheduling queue dynamicallyas shown in Algorithm 2

62 Algorithm Complexity There is only one loop in Algo-rithm 2 that is the 7th step is the loop of the length that isequal to the number of tasks in the ready queue Thus thecomputational complexity of the Algorithm 2 is 119874(119899 times 119898)and the computational complexity of the whole schedulingalgorithm is 119874(119899 times 119898)


120579119894 preemption threshold of 119879119894119901119903119894 priority of 119879119894119878119895 ready tasks queue 1 le 119895 le 119899119906119898119899119906119898 is the number of tasks in the queue119879119886119903119903 task just arriving119879119890119909119890 task in executing119879119898119886119909 the highest priority task of 1198781198951 119879119886119903119903 arrive2 IF 119899119906119898 = 03 119879119890119909119890 = 1198791198861199031199034 EXECUTE 1198791198901199091198905 ELSE6 Compute 119901119903119886119903119903 using (7)7 FOR 119894 = 1 to 119899119906119898 DO8 Compute 119901119903119894 using (5)9 Compute 12057911989410 SORT task in 119878119895 by 119901119903119895 Desc11 END IF12 IF 119901119903119886119903119903 gt 12057911989811988611990913 119879119890119909119890 = 11987911988611990311990314 IF 119879119894119899 in 119878119895 canrsquot Finish15 ABORT 11987911989411989916 ELSE17 119879119890119909119890 = 11987911989811988611990918 ADD 119879119886119903119903 to 11987811989519 119899119906119898 + +20 END IF21 EXECUTE 11987911989011990911989022 END IF

Algorithm 2 Task scheduling

7 Simulations and Analysis

71 Simulation Experiments The experiment platformrsquos CPUis Intel dual-core 28GHZprocessor and thememory is 8GBThe operating system is Ubuntu Linux and the source codeis implemented by C language

In the experiment we consider the aperiodic soft real-time task Any task119879119894rsquos arrival time 119886119894 obeys Poisson distribu-tion (120582 = 4) and the estimated execution time 119890119905119894 is subject to[30 50] average distribution Therefore its relative deadline119889119894 is [12 15] times 119890119905119894 average distribution its critical time 119888119903119894 is[12 15] times 119889119894 and its value 119881119894 is randomly selected between[10 50]

Performance indicators are included as follows

(1) Systematic success rate (SSR)

119878119878119877 = 119873119904119873 (8)

where 119873119904 denotes as the number of successful tasksand119873 represents the number of system overages

(2) Average task latency rate ATDR

119860119879119863119877 = 119886V119892( sum119879119894isin119879119878

119890119897119894119890119905119894) (9)

where 119879119878 denotes the successful task set and 119890119897119894 and119890119905119894 denote execution latency and execution time of thesuccess task 119879119894

(3) Effective utilization Rate (EUR)

119864119880119877 = 119862119879119904119862119879 (10)

where 119862119879119904 denotes as the CPU time for successfultasks and 119862119879 represents the total CPU time inscheduling

(4) Cumulative Value (CV)

119862119881 = sum119879119894isin119879119878

119881119894 (11)

where 119879119878 denotes the successful task set and repre-sents the value of the task 119879119894

72 PerformanceAnalysis Under soft real-time environmentthe preemptive threshold scheduling strategy is comparedwith Earliest Deadline First (EDF) method and Least SlackFirst (LSF) method and the simulation results and perfor-mance are analyzed as follows

(1) Figure 2 shows simulation results for SSR scheduled byDSPT EDF and LSF The taskrsquos arrival interval is representedon the horizontal axis and the vertical axis is SSR Whenthe arrival interval of the task is very short the preemptionamong the tasks causes the SSR to be lowWith the increasingof arrival interval the preemption among tasks decreasesand the SSR arguments From Figure 2 SSR is sorted indescending order followed by DSPT LSF and EDF This isdue to the EDF algorithm chooses task with earliest deadlinebut not most urgent which can lead to many tasks failWhile the LSF algorithm always chose the task with theshortest slack time dynamically its unrestricted preemptioncauses the thrashing which can make lots of task fail TheDSPT can adapt to the system environment dynamically andits preemption threshold can avoid the invalid preemptionamong tasks (2)The simulation results of ATDRare shown inFigure 3 in which the ATDR is on the vertical axis and arrivalinterval is on the horizontal axis It can be seen from Figure 3the ATDR decreases since the arrival interval grows from 25while ATDR is lower because the intervals among tasks are soshort that many of these are dropped to 10 Moreover EDFalgorithm leads to the highestATDRwhile theDSPTmethodproduces the lowest ATDR Because the EDF algorithmalways chooses the task with earliest deadline of slack time toexecute many tasks miss the deadline and delay Relative tothe EDFmethod the LSF algorithm assigns themost urgencytask to execute in time But its unconstrained preemptionmaycause serious thrashing among tasks under system overloadand this increases the ATDR The DSPT approach choosesthe most urgency task to execute and limits the preemptionamong the tasks by the threshold thereby reduces the delayof the task effectively

(3) The simulation results for EUR scheduled by EDFLSF and DSPT are shown in Figure 4 where the horizontal


EDFLSFDSPT

25 4 55 7 851Arrival Time Interval

02

04

06

08

10

SSR

Figure 2 SSR results

EDFLSFDSPT

00

01

02

03

04

ATD

R


Figure 3 ATDR results

axis is taskrsquos arrival interval and the vertical axis is EUR Theshort arrival interval means intense preemption among taskswhich causes the EUR lowWhile the arrival interval becomeslonger there are EUR arguments as the tasksrsquo preemptiondecreases From Figure 4 DSPT obtains the best EUR andEDF has the least EUR Because the EDF algorithm alwaysexecute the task with earliest deadline which causes moreinvalid CPU utilization than LSF along with the task failureThe DSPT using preemption threshold can improve the EURthrough reducing invalid preemption of tasks

(4) Figure 5 plots the systemrsquosCVas arrival interval growsIt can be seen from Figure 5 the CV increases when thearrival interval of tasks increases DSPT obtains the highestCV because it assigns higher executing priority of the taskwith high value density The LSF method produces betterCV than EDF method because under the same system loadthat more tasks finish augments CV obtained with the formalapproach

EDFLSFDSPT


02

04

06

08

10

EUR

Figure 4 EUR results

EDFLSFDSPT

1000

2000

3000

4000

5000

6000

7000

8000

CV


Figure 5 CV results

8 Conclusions and Future Works

In this paper we consider video streaming scheduling inwireless networks as soft real-time scheduling The soft real-time task model has been constructed to express streamingpacket and its quality measure as value and urgency metricare explored in scheduling Response time analysis (RTA)approach has been used to test the taskrsquos schedulability andpreemption threshold has been added to the schedule torelease the invalid preemptionThe scheduling schemaDSPTcan constrain the delay and improve the success ratio andbenefit effectively

In future works we want to apply our scheduling methodto practical wireless networks like 5G net to test its practicaleffect Moreover the stream scheduling of mixed hard real-time and soft real-time characterizes are also a study work inthe next step


Data Availability

The data including taskrsquos properties and performance indi-cators in the experiments used to support the findings ofthis study are available from the corresponding author uponrequest

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was supported by the National Natural Sci-ence Foundation of China (NSFC) (Grant nos 6156204441661083) and the Science and Technology Research Projectof Jiangxi Provincial Department of Education (Grant nosGJJ170234 GJJ160781)

References

[1] L De Cicco and S Mascolo ldquoAn adaptive video streaming con-trol systemModeling validation and performance evaluationrdquoIEEEACM Transactions on Networking vol 22 no 2 pp 526ndash539 2014

[2] O Oyman and S Singh ldquoQuality of experience for HTTPadaptive streaming servicesrdquo IEEE Communications Magazinevol 50 no 4 pp 20ndash27 2012

[3] T C Thang H T Le A T Pham and Y M Ro ldquoAn evaluationof bitrate adaptation methods for HTTP live streamingrdquo IEEEJournal on Selected Areas in Communications vol 32 no 4 pp693ndash705 2014

[4] L He and F Li ldquoContent and buffer status aware packetscheduling and resource management framework for videostreaming over LTE systemrdquoEurasip Journal on Image andVideoProcessing vol 2017 no 1 2017

[5] AM Sheikh A Fiandrotti and EMagli ldquoDistributed schedul-ing for scalable P2P video streaming with network codingrdquoin Proceedings of the 32nd IEEE Conference on ComputerCommunications (IEEE INFOCOM rsquo13) pp 11-12 April 2013

[6] AM Sheikh A Fiandrotti and EMagli ldquoDistributed schedul-ing for low-delay and loss-resilient media streaming withnetwork codingrdquo IEEE Transactions on Multimedia vol 16 no8 pp 2294ndash2306 2014

[7] N Thomos E Kurdoglu P Frossard and M Van Der SchaarldquoAdaptive prioritized random linear coding and scheduling forlayered data delivery from multiple serversrdquo IEEE Transactionson Multimedia vol 17 no 6 pp 893ndash906 2015

[8] S Huang E Izquierdo and P Hao ldquoAdaptive packet schedulingfor scalable video streaming with network codingrdquo Journal ofVisual Communication and Image Representation vol 43 pp10ndash20 2017

[9] M Zhao X Gong J Liang W Wang X Que and S ChengldquoQoE-driven cross-layer optimization for wireless dynamicadaptive streaming of scalable videos over HTTPrdquo IEEE Trans-actions on Circuits and Systems for Video Technology vol 25 no3 pp 451ndash465 2015

[10] D-R Chen ldquoA real-time streaming control for quality-of-service coexisting wireless body area networksrdquo Applied SoftComputing 2017

[11] Y Cao Q Liu Y Zuo G Luo H Wang and M HuangldquoReceiver-assisted cellularwifi handover management for effi-cient multipath multimedia delivery in heterogeneous wirelessnetworksrdquo EURASIP Journal on Wireless Communications andNetworking vol 2016 no 1 2016

[12] C Sieber T Hosfeld T Zinner P Tran-Gia and C Tim-merer ldquoImplementation and user-centric comparison of a noveladaptation logic for DASH with SVCrdquo in Proceedings of the2013 IFIPIEEE International Symposium on Integrated NetworkManagement IM 2013 pp 1318ndash1323 Belgium May 2013

[13] S A Hosseini F Fund and S S Panwar ldquo(Not) yet anotherpolicy for scalable video delivery tomobile usersrdquo inProceedingsof the 7th ACMWorkshop on Mobile Video MoVid 2015 pp 17ndash22 USA March 2015

[14] S A Hosseini and S S Panwar ldquoRestless streaming banditsScheduling scalable video in wireless networksrdquo in Proceedingsof the 2017 55th Annual Allerton Conference on CommunicationControl and Computing (Allerton) pp 620ndash628 Monticello ILUSA October 2017

[15] Y-H Yeh Y-C Lai Y-H Chen and C-N Lai ldquoA bandwidthallocation algorithm with channel quality and QoS aware forIEEE 80216 base stationsrdquo International Journal of Communi-cation Systems vol 27 no 10 pp 1601ndash1615 2014

[16] J Kim and E-S Ryu ldquoQoS optimal real-time video streaming indistributed wireless image-sensing platformsrdquo Journal of Real-Time Image Processing vol 13 no 3 pp 547ndash556 2017

[17] M Xing S Xiang and L Cai ldquoA real-time adaptive algorithmfor video streaming over multiple wireless access networksrdquoIEEE Journal on Selected Areas in Communications vol 32 no4 pp 795ndash805 2014

[18] H R Mendis N C Audsley and L S Indrusiak ldquoTaskallocation for decoding multiple hard real-time video streamson homogeneous NoCsrdquo in Proceedings of the 13th InternationalConference on Industrial Informatics INDIN 2015 pp 246ndash251UK July 2015

[19] Y Cao F Song Q Liu M Huang H Wang and I You ldquoALDDoS-Aware energy-efficientmultipathing scheme formobilecloud computing systemsrdquo IEEE Access vol 5 pp 21862ndash218722017

[20] P Dziurzanski A K Singh and L S Indrusiak ldquoFeedback-based admission control for hard real-time task allocationunder dynamic workload on many-core systemsrdquo in Architec-ture of computing systemsmdashARCS 2016 vol 9637 of LectureNotes in Comput Sci pp 157ndash169 Springer [Cham] 2016

[21] W Wang Y Cao J Gong and Z Li ldquoCP-TPS A Real-TimeTransaction Processing Strategy Supporting CompensatoryTaskrdquo in Proceedings of the 20th IEEE International Conferenceon Computational Science and Engineering and 15th IEEEIFIPInternational Conference on Embedded andUbiquitous Comput-ing CSE and EUC 2017 pp 475ndash482 China July 2017

[22] H Mendis N Audsley L Indrusiak et al ldquoDynamic and statictask allocation for hard real-time video stream decoding onNoCsrdquo Leibniz Transactions on Embedded Systems vol 4 no2 pp 01ndash25 2017

[23] JWuC YuenN-MCheung and J Chen ldquoDelay-ConstrainedHigh Definition Video Transmission in Heterogeneous Wire-less Networks with Multi-Homed Terminalsrdquo IEEE Transac-tions on Mobile Computing vol 15 no 3 pp 641ndash655 2016

[24] JWu C Yuen B ChengMWang and J Chen ldquoAdaptive FlowAssignment andPacket Scheduling forDelay-ConstrainedTraf-fic over Heterogeneous Wireless Networksrdquo IEEE Transactionson Vehicular Technology vol 65 no 10 pp 8781ndash8787 2016

International Journal of

AerospaceEngineeringHindawiwwwhindawicom Volume 2018

RoboticsJournal of

Hindawiwwwhindawicom Volume 2018


Active and Passive Electronic Components

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawiwwwhindawicom

Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Control Scienceand Engineering

Journal of



Journal ofEngineeringVolume 2018

SensorsJournal of



RotatingMachinery


Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation


Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia

Submit your manuscripts atwwwhindawicom


In this paper we only consider the CPU in the systemscheduling and ignore the time caused by context switchingWe study the video streaming transmission problem ofgateway devices in wireless network at the specific level ofpackets of transport tiers

Organization The structure of the paper is organizedas follows Section 2 describes the related work Section 3designs the structure of DSPT strategy Section 4 definesthe task model and problem analysis in detail Section 5 ispriority and preemption thresholds description The detailsof scheduling based on preemption threshold are introducedin Section 6 The simulation setup and results are givenin Section 7 Finally the conclusions and future work arediscussed

2 Related Work

In wireless networks the scheduling strategy should ensurehigh quality For this purpose lots of studies were proposedThe works in [1 2] enabled high-quality HTTP streamingin which video pre-encoded into segments that can onlystart to play if all the packet have received However muchpacket lost for video application in the above work Theliterature [3] optimized the dynamic HTTP live streamingservice by segment adaption Some literatures proposedvideo streaming scheduling using video coding with networkcoding For example the literatures [5 6] optimized videostreaming by combining video coding and network codingfor P2P network and distributed network respectively How-ever it might suffer from bandwidth inefficiencies becauseof unrecoverable packets The work [7] proposed streamingscheduling for multiple server by Markov decision processesbut it had high computational complexity In [9] the authorsdesigned a crosslayer optimization scheme for dynamicwireless video streaming However these methods cannotcontain the packet urgency for streaming control

With the emergence of various networks the researchon heterogeneous wireless networks had been motivatedThe literatures [11 19] studied the concurrent multipathtransfer of the capacity-limited heterogeneous wireless plat-form Another work [17] proposed a real-time adaptive videostreaming delivermethod overmultiple wireless networks In[23 24] delay-constrained video transmission with multipleinterface in heterogeneous network was proposed

Some researchers considered video streaming as hardreal-time that is the streaming packet missing the deadlinewill be considered invalid For example the literature [20]proposed a feedback-based control approach under hardreal-time workload to improve admission rate In literatures[18 21] the author studied in hard real-time video streamallocation to improve the utilization of video decodingNevertheless most video streaming data meets the charac-teristics of soft real-time task so delay-constrained of videois considered acceptable in applications

3 DSPT Detail Design

We consider video streaming deliver over wireless net-works as soft real-time scheduling The streaming packet

constructing priority function

middot preemption threshold selection middot response time analysis

scheduling based on preemption threshold

task model and analysis middot value density analysis

middot urgency analysis

task schedulability analysis Scheduler

execute

1 i

i

j

Figure 1 The flow diagram of DSPT strategy

is expressed as task and its quality and urgency are themetrics in scheduling The taskrsquos delay will affect the qualityTherefore we introduced the value to measure the qualityResponse time analysis (RTA) is used to test the schedula-bility of real-time task and the preemption threshold-basedscheduling approach is to reduce invalid preemption amongtasks The scheduling algorithm can effectively reduce thedelay and improve the success rate and benefit The flowdiagram of DSPT strategy is shown in Figure 1 In this paperwe assume that the value function 119881119894(119905) of the soft real-time task 119879119894 keeps unchanged in the interval [119886119894 119889119894) anddecreases linearly in the interval While executing in theinterval [119889119894 119888119903119894) the task 119879119894 is not allowed to be preemptedin order to protect the execution of the task

Contribution 1 We construct the priority assignment func-tion combining value density and urgency improving thesuccess and quality of task soft real-time scheduling

Contribution 2 We propose the task scheduling strategyDSPT based on preemption threshold which can obtainbetter performance

4 Task Model and Analysis

Considering video streaming with delay-constrained underwireless we conduct soft real-time task model in this section

41 Soft Real-Time Task Model We refer to a task as anaperiodic task model and the task is soft real-time Thissection provides the definitions of soft real-time task modelused in the paper

Task 119879119894 An aperiodic task 119879119894 is defined as a triple 119879119894 =119886119894 119890119905119894 119889119894 119888119903119894 where 119886119894119890119905119894119889119894 and 119888119903119894 are the arrival timethe estimated execution time the relative deadline and thecritical time of task 119879119894

Having executed time ℎ119905119894 ℎ119905119894 denotes that the task hasbeen executing ℎ119905119894 units at time 1199050

Required execution time 119903119905119894 119903119905119894 denotes the remainingexecution time of task 119879119894 Thus it meets 119903119905119894 = 119890119905119894 minus ℎ119905119894

Executable time 119897119905119894 119897119905119894 is the executable time of the task 119879119894prior to its deadline 119889119894 where at the time 1199050(1199050 lt 119889119894)












119881119863119894 (119905) =


119881119894119903119905119894 119889119894 le 119905 lt 119888119903119894 (2)














119901119903119894 = 119880119892119894 (119905) times ln [1 + 119881119863119894 (119905)] (4)














119877119905119894 = 119890119905119894 + sum119879119895isin119864119869119878119894

119903119905119895 (7)






















119878119878119877 = 119873119904119873 (8)



119860119879119863119877 = 119886V119892( sum119879119894isin119879119878

119890119897119894119890119905119894) (9)



119864119880119877 = 119862119879119904119862119879 (10)



119862119881 = sum119879119894isin119879119878

119881119894 (11)






EDFLSFDSPT


02

04

06

08

10

SSR


EDFLSFDSPT

00

01

02

03

04

ATD

R





EDFLSFDSPT


02

04

06

08

10

EUR


EDFLSFDSPT

1000

2000

3000

4000

5000

6000

7000

8000

CV


Figure 5 CV results





Data Availability




Acknowledgments


References



























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia













119881119863119894 (119905) =


119881119894119903119905119894 119889119894 le 119905 lt 119888119903119894 (2)














119901119903119894 = 119880119892119894 (119905) times ln [1 + 119881119863119894 (119905)] (4)














119877119905119894 = 119890119905119894 + sum119879119895isin119864119869119878119894

119903119905119895 (7)






















119878119878119877 = 119873119904119873 (8)



119860119879119863119877 = 119886V119892( sum119879119894isin119879119878

119890119897119894119890119905119894) (9)



119864119880119877 = 119862119879119904119862119879 (10)



119862119881 = sum119879119894isin119879119878

119881119894 (11)






EDFLSFDSPT


02

04

06

08

10

SSR


EDFLSFDSPT

00

01

02

03

04

ATD

R





EDFLSFDSPT


02

04

06

08

10

EUR


EDFLSFDSPT

1000

2000

3000

4000

5000

6000

7000

8000

CV


Figure 5 CV results





Data Availability




Acknowledgments


References



























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia













119877119905119894 = 119890119905119894 + sum119879119895isin119864119869119878119894

119903119905119895 (7)






















119878119878119877 = 119873119904119873 (8)



119860119879119863119877 = 119886V119892( sum119879119894isin119879119878

119890119897119894119890119905119894) (9)



119864119880119877 = 119862119879119904119862119879 (10)



119862119881 = sum119879119894isin119879119878

119881119894 (11)






EDFLSFDSPT


02

04

06

08

10

SSR


EDFLSFDSPT

00

01

02

03

04

ATD

R





EDFLSFDSPT


02

04

06

08

10

EUR


EDFLSFDSPT

1000

2000

3000

4000

5000

6000

7000

8000

CV


Figure 5 CV results





Data Availability




Acknowledgments


References



























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia










119878119878119877 = 119873119904119873 (8)



119860119879119863119877 = 119886V119892( sum119879119894isin119879119878

119890119897119894119890119905119894) (9)



119864119880119877 = 119862119879119904119862119879 (10)



119862119881 = sum119879119894isin119879119878

119881119894 (11)






EDFLSFDSPT


02

04

06

08

10

SSR


EDFLSFDSPT

00

01

02

03

04

ATD

R





EDFLSFDSPT


02

04

06

08

10

EUR


EDFLSFDSPT

1000

2000

3000

4000

5000

6000

7000

8000

CV


Figure 5 CV results





Data Availability




Acknowledgments


References



























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



EDFLSFDSPT


02

04

06

08

10

SSR


EDFLSFDSPT

00

01

02

03

04

ATD

R





EDFLSFDSPT


02

04

06

08

10

EUR


EDFLSFDSPT

1000

2000

3000

4000

5000

6000

7000

8000

CV


Figure 5 CV results





Data Availability




Acknowledgments


References



























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



Data Availability




Acknowledgments


References



























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia




RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia


Documents

Dynamic Soft Real-Time Scheduling with Preemption ...downloads.hindawi.com/journals/ijdmb/2019/5284968.pdf · works as so real-time scheduling. e streaming packet constructing priority